Understanding the effects of tsunami-induced debris flows on road infrastructure is essential for developing effective mitigation strategies in coastal regions. This study develops fragility curves for the flexible pavement road network in Karachi, Pakistan, using a hybrid dataset comprising observed post-tsunami damage and synthetically generated samples to address class imbalance. Data augmentation improved the predictive performance of the machine learning-based model significantly, increasing severe damage classification accuracy from near-zero to nearly 97%. Logistic regression was used to model the probability of exceeding three damage levels (DL1–DL3) based on inundation depth. The resulting fragility curves captured a clear progression in damage severity, with confidence intervals indicating high reliability at mid-depth ranges and greater uncertainty at extreme values. Model accuracy evaluation (98.66%) and goodness-of-fit metrics supported model validity, with root mean square error ranging from 0.0701 to 0.1061 and McFadden’s R2 values from 0.0237 (DL3) to 0.1724 (DL1), confirming smooth and consistent probability transitions across damage states. These regression-based fragility models offer a statistically robust and practical framework for assessing tsunami-induced road vulnerability, contributing valuable insights for disaster preparedness and infrastructure resilience planning in high-risk coastal environments.
Rafi et al. (2026) studied this question.